Abstract:When multispectral images are generated by mosaicking and registration, they are prone to strip noise and mismatch due to the effects of clouds and water. In the paper, a combined method of images registration based on SIFT algorithm constraints to 3D information and multispectral images preprocessing algorithm under the restriction of ground spectral characteristics is proposed. In this algorithm, SIFT operator constraints to 3D information are used to improve the registration accuracy. At the same time, effective radiation calibration points are selected under the ground spectral characteristics constraints to improve the gray-scale correction accuracy of each band image when stitching. The experimental results show that, when the method is used to preprocess the multispectral camera data, even if there are special areas such as clouds and water on the image, the images can own high registration accuracy without strip noise, and the spectral consistency is high.